What Is Meta Muse? From Flight and Hotel Comparison to Price Monitoring, an AI Agent That Actually Does the Work

In September 2026, Meta launched a new personal AI agent called Muse. The biggest difference from chatbots we already know, such as ChatGPT and Claude, is not that it “answers questions more cleverly.” It can actually carry out tasks for you.

With a typical AI, you ask a question, it gives an answer, and you still have to search websites, compare prices, fill in forms, and finish the booking yourself. Muse works differently: you tell it the end goal, and it can open a browser, visit pages, fill in forms, connect third-party apps, and run a multi-step task. Even if you close Muse, it can keep working in the background until the task is done, conditions change, or it needs your approval for the next step. Meta therefore calls Muse a Personal AI Agent.

If you travel often, collect points and miles, or hunt for deals, that shift is interesting. The time sink is usually not “looking up one answer.” It is searching, comparing, and monitoring the same thing over and over.

Why is Muse different from ordinary AI?

A key piece of Muse is the Muse Secure VM. In plain terms, every Muse has its own virtual computer running in the cloud, with its own browser and environment. Through that machine, Muse can browse sites, fill in forms, handle customer service, book travel, and keep going after you leave the app.

So instead of asking:

Which day in November is cheapest for Vancouver to Tokyo?

A Muse-style instruction looks more like this:

Search Vancouver–Tokyo flights in November. Dates can flex five days either side, and compare airlines and nearby airports. If you find a clearly cheaper business-class option, give me the top three itineraries.

Muse can grind through the repeated searches itself, instead of only telling you where to look.

It can also connect third-party services such as Email and Calendar, and act on them with your permission. Sensitive steps such as sending email or buying something need your approval. Inside the Secure VM there is also a Sentinel, isolated from the main agent, that checks what Muse does with the internet. You can decide whether each connected service is read-only, or whether Muse may take further actions.

Use Muse to search and compare flights and hotels

For frequent travellers, one of the easiest uses to picture is messy search.

A simple flight search may need one set of dates. Once dates are flexible, it gets ugly fast. You might leave from YVR, SEA, SFO, or LAX, fly to Tokyo or Osaka, shift dates by a few days, and then mix economy, business, and different connections. The combinations can run into the dozens or hundreds.

That is the kind of work an AI agent is built for, because it will not lose patience after a dozen searches the way a person does. You can tell Muse:

Search business class from Vancouver to Tokyo over the next three months. Dates can be flexible. Compare departures from YVR, SEA, SFO, and LAX, and keep only itineraries with a reasonable total travel time. Give me the five most worth considering, with price, airline, routing, and change/cancellation rules.

Hotels work the same way. You can have Muse weigh price, location, ratings, room type, and cancellation policy together, instead of simply hunting for “the cheapest hotel.” Meta’s official materials also list booking travel as a task Muse can do.

For points collectors, the more interesting move is putting cash fares and award tickets side by side: ask Muse to check paid business class plus programs such as Aeroplan, Flying Blue, and Avios, then work out whether using points is actually a good deal.

One caveat: there is no complete public success-rate data yet on whether Muse can query airline and loyalty-program sites reliably over time. Complex award searches are well worth trying, but you still have to test the results yourself.

More useful than search: watching flights and hotels you already booked

For frequent travellers, I think Muse’s more valuable job is not “book me a flight once.” It is watching travel you already booked, for a long stretch.

Plenty of hotels allow free cancellation. If you booked a room three months ago at CAD 450 a night, and two months later the same room type is CAD 360, you can theoretically cancel and rebook as long as the original reservation is still free to cancel.

The catch is that most people will not re-search a hotel they already booked, every day.

Muse’s official page lists monitoring prices as a task a personal agent can do, and Muse can keep running after you close the app. So you can give it a long-running job:

Monitor all of my upcoming hotel reservations. If a lower price appears for the exact same dates, room type, and cancellation terms, calculate how much I would actually save by rebooking. Do not cancel anything automatically. Notify me first.

Flights follow the same idea. Muse can watch fares, schedules, or better alternatives. If the fare drops meaningfully, it can then check the original ticket’s change rules and whether reissuing would actually save money.

That is the difference between an agent and ordinary AI: we used to ask for a price once. Now we can have the AI keep watching the price for us.

The same idea works for product prices

This is not only for travel. If you want a computer but think the current price is too high, you can tell Muse:

Monitor this computer. If any reliable seller’s price drops below $1,500 in the next month, notify me, but do not buy it automatically.

Muse keeps checking in the background and comes back when the condition is met. Meta also lists product search, online purchases, and price monitoring as important Muse uses, and it partners with Stripe Link for payment. Eligible orders can get purchase protections that include price-drop protection.

AI agents are also hitting site-side limits. Amazon, for example, has started blocking some of Muse’s automated shopping on its platform, so “let one agent roam every website” is not reality yet.

Muse can also handle the chores we keep putting off

After Muse launched, the social-media cases that got the most attention were not fancy AI reasoning. They were things people already knew they should do, and kept putting off because they were annoying.

Users asked Muse to:

• Compare auto-insurance quotes
• Request a refund
• Cancel a subscription they no longer use
• Contact support about a bill
• Track a product price drop
• File a claim after a delayed or cancelled flight
• Find a cheaper mobile plan
• Handle a return

The dollar amounts in those stories mostly come from users themselves, so treat them as unverified personal anecdotes, not a result every Muse user should expect. They still make the point: the most useful part of an AI agent may not be doing what we cannot do. It is doing the chores we can do, but that are not worth our time. If you want more examples, see the Use of Muse site.

Can Muse pay and book for me?

In some cases, yes — but it is not unsupervised auto-spend. Meta says Muse can complete purchases and pay through Stripe Link with one-time card numbers, so your real card details are not handed straight to the merchant or to Muse. For sensitive steps such as sending email or completing a purchase, Muse asks you to confirm first and keeps a full activity log.

So the more sensible setup is not:

Give Muse my credit card and let it buy cheap flights on its own.

It is:

When you find a flight that meets my conditions, notify me with the price, rules, and savings. Execute only after I approve.

For travel, points transfers, and large purchases, the AI executes, a human approves is clearly the safer pattern.

Privacy and security still matter

The more an agent can do, the more data it needs. If you want Muse to spot flight and hotel bookings automatically, it may need Email. If you want it to coordinate travel, it may need Calendar. Shopping involves payment and account logins.

Meta says Muse login credentials sit in a separate secure store, and the agent itself cannot read plaintext passwords. Muse conversations and Secure VM data are not fed to Meta’s ad system, and you can turn off using your interactions to train Meta AI models. Meta also plans a Muse Confidential VM, so the whole virtual machine is encrypted with a key only you hold.

At the same time, Muse is still very new. Reuters reported that Meta found security and reliability issues in internal testing. A phone-support feature Muse recently tested may even involve human contractors on some calls, which raises fresh privacy questions.

A reasonable way to start is low-risk work: price search, public-web research, and travel comparison. Open Email, Calendar, shopping, and payment permissions only after you are comfortable.

How do you use Muse today?

Muse launched on 8 September 2026. You can use it on iOS, Android, the Muse web app, and WhatsApp, with Meta’s AI Glasses planned later.

Download the app or use the web version. Within 48 hours of opening an account, enter my invite code 35HBVP in Settings → General. Both sides get 1 billion tokens.

The free tier has usage limits. Heavier users can subscribe. Reuters reported paid plans at $20 and $100 a month.

Growth after launch has been fast. On 22 September, Reuters reported more than 2.5 million downloads about two weeks after release, and a high ranking on the US App Store.

Takeaway: from “look this up for me” to “keep watching it”

What is interesting about Muse is not that Meta built another chatbot. It is that how we use AI is changing.

That is a shift from one-off Search to Search + Monitoring + Action.

For frequent travellers, points collectors, and people who like working a deal, that is the part of Muse worth watching. We do not lack an AI that can tell us where to look up a price. We lack an assistant that will search dozens of times, watch a price for months, and only tap us when it is actually worth acting.

Appendix: 10 sample prompts

Let Muse run search and monitoring on its own. Anything involving payment, cancelling a booking, changing a ticket, or transferring points should still require your confirmation. ⚠️ Note: these examples are only starting points. Prompts do not have to be written this way, and the list is fairly narrow. For more use cases, see the site mentioned above: Use of Muse

1. Monitor price drops on booked flights

Monitor all of my currently booked flights. For each reservation, regularly check whether the exact same flight, cabin, fare type, or a clearly better equivalent becomes cheaper. If the price drops, calculate the real savings after considering change fees, cancellation rules, travel credits, and any fare restrictions. Do not cancel, change, or rebook anything without my approval. Only notify me when the savings are meaningful.

2. Monitor price drops on booked hotels

Monitor all of my upcoming hotel reservations. Check whether the same hotel, room type, dates, and cancellation conditions become cheaper. Also check whether a better room, rate, package, or member rate becomes available at a similar total price. If a better option appears, compare it with my current reservation and tell me the exact savings or improvement. Do not cancel or rebook anything without my approval.

3. Flexible-date search for the cheapest flights

Find the best flight options from Vancouver to Tokyo. My travel dates are flexible within a 10-day window around my preferred departure date. Compare multiple departure dates, return dates, nearby airports, airlines, and reasonable connection options. Show me the cheapest economy, premium economy, and business class options. Prioritize good total travel time and avoid unreasonable overnight connections. Give me the top 5 options with total price, airline, routing, travel time, fare conditions, and booking source.

4. Compare cash fares and award tickets

Search for both cash fares and award availability for my trip. Compare:

• Cash tickets
• Aeroplan
• Flying Blue
• Alaska Mileage Plan
• British Airways Avios
• Any other major transferable-points program that may offer better value

For award tickets, include:

• Miles or points required
• Taxes and fees
• Transfer partners
• Transfer ratios
• Whether availability is actually bookable

Calculate an approximate cents-per-point value compared with the best reasonable cash fare. Do not recommend transferring points until award space is confirmed.

5. Keep looking for business-class awards

Monitor business-class award availability for the following trip:

Origin airports: YVR, SEA, SFO, LAX

Destination: Tokyo

Travel window: March 10 to March 20

Passengers: 2

Programs to prioritize: Aeroplan, Alaska Mileage Plan, Flying Blue, Avios

Notify me only when you find at least 2 business-class seats on the same flight. Prioritize nonstop flights first, then one-stop itineraries. Include the mileage cost, taxes, transfer partners, cancellation rules, and whether the seats are actually bookable.

6. Automatically find award sweet spots

Search for unusually good business-class award opportunities departing from Vancouver over the next 6 months.

I am flexible on destination. Focus on:

• Asia
• Europe
• Hawaii
• Australia and New Zealand

Look for awards that are meaningfully cheaper than normal pricing or offer unusually high value.

Compare Aeroplan, Flying Blue, Alaska Mileage Plan, Avios, and other relevant programs.

Only show opportunities that are realistically bookable.

Rank the results by:

1. Award value
2. Availability
3. Ease of booking
4. Total taxes and fees

7. Monitor flight times and schedule changes

Monitor all of my booked flights for schedule changes.

Notify me if any of the following happens:

• Departure or arrival time changes by more than 30 minutes
• Aircraft changes
• Cabin changes
• Connection time becomes too short
• Flight number changes
• A nonstop flight becomes a connecting itinerary
• My seat assignment disappears
• The airline changes or cancels the flight

If a major schedule change occurs, check whether better alternative flights are available before notifying me.

Do not make any changes without my approval.

8. Compare hotel cash rates and points rates

For my upcoming trip, compare hotel cash rates and points rates.

For each hotel, calculate:

• Cash price including taxes and mandatory fees
• Points required
• Approximate cents-per-point value
• Resort fees
• Parking fees
• Breakfast benefits
• Elite-status benefits
• Cancellation policy

Prioritize hotels where using points provides unusually good value.

Also check whether transferring bank points to a hotel program would make sense.

Do not transfer any points or make a reservation without my approval.

9. Monitor hotel award-night availability

Monitor award availability at the following hotels for my dates.

I need:

• 2 guests
• Standard room
• Points booking
• Flexible cancellation if possible

Check regularly for newly released award nights.

If the hotel becomes available on points, notify me immediately and include:

• Points required per night
• Cash price
• Approximate redemption value
• Cancellation policy
• Whether elite benefits apply

Do not book anything without my approval.

10. Full travel price-monitoring agent

Act as my personal travel price-monitoring assistant.

For all upcoming trips in my calendar and email, identify:

• Flights
• Hotels
• Rental cars

For each booking, monitor:

• Price drops
• Better refundable rates
• Better award availability
• Better cabin or room options
• Schedule changes
• Better routing
• Promotions that could reduce my total cost

Only notify me when there is a meaningful opportunity or important change.

When you notify me, clearly show:

1. My current booking
2. The new option
3. The exact savings or improvement
4. Any risks, fees, or restrictions
5. What action you recommend I consider

Never cancel, change, rebook, transfer points, or make a purchase without my approval.